I don't know why there aren't more scopes of that kind. Maybe because this renders HF measurements more difficult.
It's definitely much more complex, even for the relatively low-spec TPS2xxx series (I think max BW is only 200MHz). They also produce more noise because of the isolation circuit. The Fluke ScopeMeters are also in that category by the way, but as far as I know, very few other manufacturers. Not sure why Agilent or Lecroy don't have one.
It's said that the clean way to measure differetials is to use a differential probe.
But I just recently saw an offer for the regular differential probe for the Agilent InfiniiVision 2000 X-series that was priced at about 1376€. Jee! that's more expensive than the entire scope!!
Yes, probes are expensive. Note that even a good, brand-name passive probe will easily exceed $100, sometimes even $300, it's not just a piece of cable. Active probes are obviously even more complex. If you don't need high bandwidth or high CMRR, I believe there are some cheaper differential probes (eg. by Testec and Picotech) with bandwidth up to 100MHz for something like $300-$500. They will obviously need an external power supply, since the probe power interface tends to be proprietary. No experience them, I expect them to be OK, although common mode rejection ratio (CMRR) at high(ish) frequencies or voltages (voltage coefficient) will probably be inferior to the expensive brands because of a cheaper attenuator design, since good CMRR requires very careful matching of the attenuator components between channels.
On the other hand, one may use 2-inputs with ground unplugged and switching to A - B mode, but that wastes one precious little input chanel. And besides, I doubt the sample quality in this mode of operation would be as good as using a true floating inputs scope for my purposes.
CMRR will be horrible, probably only 20:1 or 50:1, even at low frequencies, as opposed to 1000:1 and better for a good diff probe. How important this is depends on your common mode signal. Don't be surprised to see a 50/60Hz sine superimposed on your signal.
Or is there a reasonable feasable way to transform a regular Agilent scope into a floating input scope without sacrificing the security of a protective earth shielded chassis and without completely compromising the EMI characteristics?
Not without spending a lot of money or time. An isolation amplifier might work (Tek made one at some point), but is not exactly cheap or simple.
I guess one would need at least a new pair of input sockets (2 signals+GND instead of 1 signal+GND each) and a new pair of probes(2 twisted signal lines shielded by GND instead of just 1 signal line shielded by GND each). Would that be more economic than buying an overfeatured and overprized differential probe?
What's the point of this? What would you connect the two signals to, the input circuitry is only single-ended. There were some scopes with differential inputs (the Tek 7000/11000 series with the correct plugins, for example), but that's fairly rare. I guess you could build your own differential probe if you don't require high bandwidth or high voltage. There is at least one design (by Elektor?) around. They usually use something like a differential video line driver as active stage, and some sort of attenuation before this. This is basically what commercial probes are doing. I wouldn't expect great bandwidth or CMRR from a DIY design, though.